[FIX] ASR: round half away from zero for window/bad-chan counts

[sappelhoff]

numpy's np.round (banker's rounding, round-half-to-even) and Python's int() truncation diverge from MATLAB's round-half-away-from-zero at exact-half inputs, giving ±1 counts for the bad-channel count and window length N in clean_windows and asr_calibrate. Add a _round_half_away helper and apply it at those sites.

Reachability is narrow (ties need specific channel counts or odd-integer sfreq with win_len=0.5; common rates never trigger), so this mainly hardens the edge cases.

Testing

Added a helper unit test covering ties (2.5→3, 128.5→129, -2.5→-3), divergence from np.round, array input, and int() exactness. Full tests/test_asr.py passes; ruff clean.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 83.11%. Comparing base (3721224) to head (0daabd0).

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #116   +/-   ##
=======================================
  Coverage   83.10%   83.11%           
=======================================
  Files          25       25           
  Lines        2818     2819    +1     
=======================================
+ Hits         2342     2343    +1     
  Misses        476      476           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[sappelhoff]

Note: includes the nonlinear_eigenspace reproducibility fix from #118. test_asr_class[*-riemann] compares incremental-vs-bulk output with a loose tolerance and is flaky on CI because that path used an unseeded RNG; this can trip on any PR regardless of what it changes. #118 is the root fix — once it merges first, this commit drops out on rebase.